(Eric R. Westervelt's Ph.D. Thesis Documentation)
Overview | Publications | Movies | Photos | MATLAB Code | Collaborators
March 2003 Experiments | July 2002 Experiments | Simulations
Eric R. Westervelt is now at The Ohio State University in the Department of Mechanical Engineering
Second RABBIT Walking Experiments (March 2003)
About these movies:
These movies, taken in March 2003, are of the biped prototype called RABBIT located at the LAG in Grenoble, France. This robot is part of the French project Commande de Robots à Pattes of the CNRS - GdR Automatique. These movies demonstrate the utility of the theoretical framework developed in the thesis. In these videos, the "training wheels" attached to the boom in no way support RABBIT's weight. The prismatic post of the training wheels simply prevents the RABBIT's hips from dropping too low (RABBIT falling to the ground) in the event of an experimental mishap.

Walking at 0.7 m/s:
Description: In this first experiment, RABBIT was controlled with a feedback designed to induce walking at 0.7 m/s. The experiment lasted approximately 93 seconds during which RABBIT took 170 steps making six laps about the center stand.

Demonstration of robustness to perturbations:
Description: This second experiment demonstrates the robustness of controllers designed via the theoretical framework. Two types of perturbations were applied to RABBIT controlled by a feedback designed to induce walking at 0.9 m/s. The first was a 10 kg mass added to the torso (the black plate shown in the videos), which resulted in a shift of the average walking rate from 0.9 m/s to 1.0 m/s. In addition to the sizable perturbation to the robot's model (RABBIT weighs 32 kg), the second perturbation was short duration forces applied to the RABBIT's torso by an experimenter in both the forward and reverse directions. Despite both these significant perturbations, RABBIT did not fall during the experiment which lasted approximately 74 seconds where RABBIT took 164 steps.

Transitioning between controllers:
Description: This third experiment demonstrates the use of transition controllers for the composition of controllers that induce walking at a fixed rate. For the experiment, the controller applied to RABBIT was transitioned between controllers at 0.1 m/s intervals from 0.5 m/s to 0.8 m/s and then back from 0.8 m/s to 0.5 m/s twice. The experiment lasted approximately 86 seconds during which RABBIT took 139 steps.

Using event-based integral control to modify the fixed point:
Description: In this fourth experiment, the same feedback used in the first experiment to induce walking at 0.7 m/s was applied with the addition of an event-based PI control used to modify the steady state average walking rate from 0.7 m/s to 0.6 m/s. The event-based control acts through step-to-step modifications of the Bézier polynomial coefficients synchronized with double support. The experiment lasted approximately 110 seconds during which RABBIT took 181 steps.

Using event-based integral control to reject a perturbation:
Description: In this fifth experiment, the same feedback used in the first experiment to induce walking at 0.7 m/s was applied but with a 10 kg mass attached to the torso. This perturbation resulted in a shift of the average walking rate from 0.7 m/s to approximately 0.85 m/s (the change in average walking rate was determined in a separate experiment not reported in the dissertation). The average walking rate of 0.7 m/s was recovered using the same event-based integral control used for the previous experiment applied on the 14th step (at approximately 11 seconds). The experiment lasted approximately 95 seconds during which RABBIT took 164 steps.

Using event-based integral control to stop the robot:
Description: In this sixth experiment, event-based integral control was used to stop RABBIT from steady state average walking rate of 0.5 m/s. This was achieved by slowing the average walking rate of RABBIT to where it did not have enough energy to successfully complete a step. The integral control used in the previous two experiments with a desired average walking rate of 0 was applied on the 34th step (at approximately 29 seconds) and RABBIT was stopped by the 39th step (at approximately 34 seconds). After stopping, RABBIT rocked back and forth until all kinetic energy from walking was dissipated.

RABBIT's joint friction:
Description: This video demonstrates RABBIT's large amount of joint friction and inertia due to its motors and gear reducers. These additional dynamics kill all internal passive motions of RABBIT.

Walking without the "training wheels":
Description: This experiment is to show that the "training wheels" that are normally attached to RABBIT's boom to provide a measure of safety (see the next experiment) do not support in any way RABBIT's weight.

RABBIT falling:
Description: In this experiment RABBIT's controller accidentally stopped causing RABBIT to fall. The experiment illustrates that 1) without feedback control RABBIT's joint friction quickly dampens out any motions and that 2) the "training wheels" attached to RABBIT's boom prevent RABBIT from falling to the ground.

Pushing RABBIT backward:
Description: This experiment demonstrates that under the feedback control developed in this work RABBIT may be pushed backward making interactions with the robot safer and more intuitive than other proposed control schemes.

Unwinding RABBIT's cables:
Description: This video demonstrates how RABBIT's cabling was unwound from the center stand after each experiment. The cabling supplies power and communications (in the form of an Ethernet cable) to the dSPACE system and power electronics located atop the center stand.

TV interview:
Description: Carlos Canudas-de-Wit and Eric Westervelt were interviewed by French Channel 3 at the conclusion of the research experiment where the above videos were taken, French Channel 3, 6:55 p.m., March 14, 2003 (MPEG-4, 13Mb).

First RABBIT Walking Experiments (July 2002)
About these movies:
These movies, taken in July 2002, are of the biped prototype called RABBIT located at the LAG in Grenoble, France. This robot is part of the French project Commande de Robots à Pattes of the CNRS - GdR Automatique. The movies below are of our preliminary tests, when RABBIT was still under construction.

First movements:
Description: In this first movie, the walking motion is executed with the robot suspended in the air. In the first part of the movie, the robot is being used a play-back device for a pre-computed walking motion. In particular, a desired steady state walking motion was pre-computed as a time trajectory and played back through the four joints via PD control. This was only done to verify correctness of controller implementation in software. In the second portion of the movie, and in ALL walking motions shown after this, no time trajectories are utilized! The robot is controlled via time-invariant, nonlinear feedback. The robot in closed loop with the controller will naturally generate its own walking motion. This is completely different than how the control of most bipeds has been realized to this day. If you don't believe this, watch the "backing up phase" of the "continuous steps movie" below. Please check out our papers for details on how this works.

First guided step:
Description: The robot has now been placed on the ground. The primary objective of this test is to make sure that the swing leg executes the planned maneuver as the robot moves forward or backward. To keep the researchers out of the workspace of the robot, ropes were attached at the robot's hips and used to move it forward and backward. (That's Jessy Grizzle and Yannick Aoustin carefully moving the robot... at this stage, they're being careful). The controlled motion may not appear to be very precise. This is because the control signals were deliberately saturated at low values to minimize any potential damage should an accident occur. This was literally the fist time the robot was supporting its own weight on the ground.

First continuous stepping:
Description: In this next movie, commutation logic has been added to the legs so that the robot swaps support leg after each impact. Jessy Grizzle and Gabriel Buche man the ropes this time. Jessy pulls on the lead rope to give the robot some initial velocity, which brings it (seemingly) into the basin of attraction of the 0.75 m/s walking motion. These is our very first attempt at walking, so Gabriel is being very cautious with the rear rope on the first test; on the second test, he already has more confidence! Because of the limited workspace available at the time these experiments were performed, the robot's motion could only hint at the asymptotic stability of the feedback controller... but it looks pretty good! In later work, we'll give the robot the ability to start and stop on his own.

Kneed Biped Walker Simulations
About these movies:
These movies are of various simulations of RABBIT performed with MATLAB. The movies were made using Geomview and using MATLAB's handle graphics.
Last modified: Mon Jul 14 16:48:08 Eastern Daylight Time 2003